Comment by qurren

4 days ago

Up next:

1. Generate a bunch of responses with both Claude and various non-Claude LLMs (ChatGPT, Gemini, Kimi)

2. Train a discriminator model that can differentiate Claude vs. non-Claude

3. Train a de-watermarking model using the discriminator model as loss

7 comments

qurren

Reply
hughw  4 days ago

Or just write a 6 line program to remove the metainfo from the file?

  • rcktmrtn  4 days ago

    Your observation is sharp, but it's not just metainfo—it's load-bearing text. To remove it, you need to delve deep and alter the tapestry of carefully selected words.

    edit: almost forgot the "—"

    • Lockal  4 days ago

      s/text/prose/ s/'/’/

      The honest seam is obvious: claudish works; imitation does not. That’s not nothing.

  • aero-glide2  4 days ago

    very difficult to remove text watermarking

    • HoldOnAMinute  4 days ago

      There must be some set of outputs that is completely resistant to watermarking. For example if I told Claude to output "Hello, World"

    • ProfessorLayton  4 days ago

      Outputs can be easily translated from the watermarked language to another, either by just knowing two+ languages, or via non-LLM translation.